Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Refactor Target Platform Capabilities Design #1276

Merged
merged 4 commits into from
Dec 2, 2024
Merged

Refactor Target Platform Capabilities Design #1276

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
2f794ad
Refactor Target Platform Capabilities Design
Nov 24, 2024
536dc92
Added tpc_platfrom_type to meta data for tp model
Nov 25, 2024
6e3bfb5
PR fixes:
Nov 27, 2024
be71127
PR fixes:
Dec 2, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions model_compression_toolkit/constants.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,6 +21,7 @@
# Metadata fields
MCT_VERSION = 'mct_version'
TPC_VERSION = 'tpc_version'
TPC_SCHEMA = 'tpc_schema'

WEIGHTS_SIGNED = True
# Minimal threshold to use for quantization ranges:
Expand Down
5 changes: 3 additions & 2 deletions model_compression_toolkit/core/common/graph/base_node.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,8 +22,9 @@
ACTIVATION_N_BITS_ATTRIBUTE, FP32_BYTES_PER_PARAMETER
from model_compression_toolkit.core.common.quantization.node_quantization_config import WeightsAttrQuantizationConfig
from model_compression_toolkit.logger import Logger
from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationConfigOptions, \
TargetPlatformCapabilities, LayerFilterParams, OpQuantizationConfig
from model_compression_toolkit.target_platform_capabilities.target_platform import TargetPlatformCapabilities, LayerFilterParams
from model_compression_toolkit.target_platform_capabilities.schema.v1 import OpQuantizationConfig, \
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Given that the schema is going to be updated (maybe frequently), and MCT core would change the schema that it is using accordingly, we need to figure out how to modify these imports throughout the code without accessing ".v1" directly at each import.

Is there a way to "export" (as TPC package "API") a default schema that references the currently used version, such that all imports will point to it and when we want to change the used schema by MCT we'll only have to change in 1 place?

Maybe @irenaby would have an idea how it can be done?

Let's discuss this offline if needed.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

BTW, this is not mandatory for this PR, but solving this here would be better, because it will save us editing all these files again in a separate PR

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The only part of MCT that should be aware of the schema is the parser once we have it. It should parse the schema into whatever representation the rest of mct works with. If we want to reuse the same classes for now, we can add a proxy module that will only import the classes from the schema, and the rest of mct imports from that proxy module.

Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I agree with @irenaby , we should have some proxy model at this stage...

QuantizationConfigOptions


class BaseNode:
Expand Down
4 changes: 2 additions & 2 deletions ...olkit/core/common/mixed_precision/resource_utilization_tools/resource_utilization_data.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,8 +23,8 @@
from model_compression_toolkit.core.common.framework_implementation import FrameworkImplementation
from model_compression_toolkit.core.common.graph.edge import EDGE_SINK_INDEX
from model_compression_toolkit.core.graph_prep_runner import graph_preparation_runner
from model_compression_toolkit.target_platform_capabilities.target_platform import TargetPlatformCapabilities, \
QuantizationConfigOptions
from model_compression_toolkit.target_platform_capabilities.target_platform import TargetPlatformCapabilities
from model_compression_toolkit.target_platform_capabilities.schema.v1 import QuantizationConfigOptions


def compute_resource_utilization_data(in_model: Any,
Expand Down
4 changes: 2 additions & 2 deletions model_compression_toolkit/core/common/quantization/candidate_node_quantization_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,8 +17,8 @@
from model_compression_toolkit.core import QuantizationConfig
from model_compression_toolkit.core.common.quantization.node_quantization_config import BaseNodeQuantizationConfig, \
NodeWeightsQuantizationConfig, NodeActivationQuantizationConfig
from model_compression_toolkit.target_platform_capabilities.target_platform import OpQuantizationConfig, \
AttributeQuantizationConfig
from model_compression_toolkit.target_platform_capabilities.schema.v1 import AttributeQuantizationConfig, \
OpQuantizationConfig
from model_compression_toolkit.logger import Logger


Expand Down
4 changes: 2 additions & 2 deletions model_compression_toolkit/core/common/quantization/node_quantization_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,8 +25,8 @@

from model_compression_toolkit.core.common.quantization.quantization_config import QuantizationConfig, \
QuantizationErrorMethod
from model_compression_toolkit.target_platform_capabilities.target_platform import OpQuantizationConfig, \
AttributeQuantizationConfig
from model_compression_toolkit.target_platform_capabilities.schema.v1 import AttributeQuantizationConfig, \
OpQuantizationConfig


##########################################
Expand Down
3 changes: 2 additions & 1 deletion ...ore/common/quantization/quantization_params_generation/qparams_activations_computation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,8 @@
import numpy as np
from typing import Dict, Union

from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod, Signedness
from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
from model_compression_toolkit.target_platform_capabilities.schema.v1 import Signedness
from model_compression_toolkit.core.common.collectors.statistics_collector import BaseStatsCollector
from model_compression_toolkit.core.common.quantization import quantization_params_generation
from model_compression_toolkit.core.common.node_prior_info import NodePriorInfo
Expand Down
2 changes: 1 addition & 1 deletion model_compression_toolkit/core/common/quantization/set_node_quantization_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@
from model_compression_toolkit.core.common.quantization.quantization_fn_selection import \
get_weights_quantization_fn
from model_compression_toolkit.target_platform_capabilities.target_platform.targetplatform2framework import TargetPlatformCapabilities
from model_compression_toolkit.target_platform_capabilities.target_platform.op_quantization_config import OpQuantizationConfig, \
from model_compression_toolkit.target_platform_capabilities.schema.v1 import OpQuantizationConfig, \
QuantizationConfigOptions


Expand Down
2 changes: 1 addition & 1 deletion ...on_toolkit/core/common/statistics_correction/apply_activation_bias_correction_to_graph.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,7 +17,7 @@
from model_compression_toolkit.core.common import BaseNode, Graph
from model_compression_toolkit.core.common.framework_implementation import FrameworkImplementation
from model_compression_toolkit.core.common.quantization.node_quantization_config import WeightsAttrQuantizationConfig
from model_compression_toolkit.target_platform_capabilities.target_platform import AttributeQuantizationConfig
from model_compression_toolkit.target_platform_capabilities.schema.v1 import AttributeQuantizationConfig


def apply_activation_bias_correction_to_graph(graph: Graph,
Expand Down
2 changes: 1 addition & 1 deletion ...l_compression_toolkit/core/common/statistics_correction/apply_bias_correction_to_graph.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,7 @@
from model_compression_toolkit.core.common import Graph, BaseNode
from model_compression_toolkit.core.common.framework_implementation import FrameworkImplementation
from model_compression_toolkit.core.common.quantization.node_quantization_config import WeightsAttrQuantizationConfig
from model_compression_toolkit.target_platform_capabilities.target_platform import AttributeQuantizationConfig
from model_compression_toolkit.target_platform_capabilities.schema.v1 import AttributeQuantizationConfig


def apply_bias_correction_to_graph(graph_to_apply_bias_correction: Graph,
Expand Down
4 changes: 2 additions & 2 deletions model_compression_toolkit/core/common/substitutions/batchnorm_reconstruction.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,8 +26,8 @@
from model_compression_toolkit.core.common.graph.base_graph import Graph
from model_compression_toolkit.core.common.graph.base_node import BaseNode
from model_compression_toolkit.core.common.graph.graph_matchers import NodeOperationMatcher
from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod, \
AttributeQuantizationConfig
from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
from model_compression_toolkit.target_platform_capabilities.schema.v1 import AttributeQuantizationConfig


class BatchNormalizationReconstruction(common.BaseSubstitution):
Expand Down
4 changes: 2 additions & 2 deletions model_compression_toolkit/core/common/substitutions/shift_negative_activation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,8 +22,8 @@
from model_compression_toolkit.core.common import FrameworkInfo, Graph, BaseNode
from model_compression_toolkit.constants import THRESHOLD, SIGNED, SHIFT_NEGATIVE_NON_LINEAR_NUM_BITS
from model_compression_toolkit.core.common.graph.graph_matchers import NodeOperationMatcher
from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod, \
AttributeQuantizationConfig
from model_compression_toolkit.target_platform_capabilities.target_platform import QuantizationMethod
from model_compression_toolkit.target_platform_capabilities.schema.v1 import AttributeQuantizationConfig
from model_compression_toolkit.core.common.quantization.set_node_quantization_config import create_node_activation_qc, \
set_quantization_configs_to_node
from model_compression_toolkit.core.common.quantization.core_config import CoreConfig
Expand Down
11 changes: 7 additions & 4 deletions model_compression_toolkit/metadata.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@

from typing import Dict, Any
from model_compression_toolkit.constants import MCT_VERSION, TPC_VERSION, OPERATORS_SCHEDULING, FUSED_NODES_MAPPING, \
CUTS, MAX_CUT, OP_ORDER, OP_RECORD, SHAPE, NODE_OUTPUT_INDEX, NODE_NAME, TOTAL_SIZE, MEM_ELEMENTS
CUTS, MAX_CUT, OP_ORDER, OP_RECORD, SHAPE, NODE_OUTPUT_INDEX, NODE_NAME, TOTAL_SIZE, MEM_ELEMENTS, TPC_SCHEMA
from model_compression_toolkit.core.common.graph.memory_graph.compute_graph_max_cut import SchedulerInfo
from model_compression_toolkit.target_platform_capabilities.target_platform import TargetPlatformCapabilities

Expand Down Expand Up @@ -43,13 +43,16 @@ def create_model_metadata(tpc: TargetPlatformCapabilities,
def get_versions_dict(tpc) -> Dict:
"""

Returns: A dictionary with TPC and MCT versions.
Returns: A dictionary with TPC, MCT and TPC-Schema versions.

"""
# imported inside to avoid circular import error
from model_compression_toolkit import __version__ as mct_version
tpc_version = f'{tpc.name}.{tpc.version}'
return {MCT_VERSION: mct_version, TPC_VERSION: tpc_version}
tpc_version = f'{tpc.tp_model.tpc_minor_version}.{tpc.tp_model.tpc_patch_version}'
lior-dikstein marked this conversation as resolved.
Show resolved Hide resolved
tpc_schema = f'{tpc.tp_model.SCHEMA_VERSION}'
return {MCT_VERSION: mct_version,
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If you need to access the fields elsewhere, why not define class or named tuple? If this goes directly into the model, no need to define global consts.

TPC_VERSION: tpc_version,
TPC_SCHEMA: tpc_schema}


def get_scheduler_metadata(scheduler_info: SchedulerInfo) -> Dict[str, Any]:
Expand Down
14 changes: 14 additions & 0 deletions model_compression_toolkit/target_platform_capabilities/schema/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,14 @@
# Copyright 2024 Sony Semiconductor Israel, Inc. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
Loading
Loading